Second-order constraints in dynamic invariant inference

Preview

Mehmet will talk about Differentiable Abstract Interpretation for Provably Robust Neural Networks by Kaituo Li, Christoph Reichenbach, Yannis Smaragdakis, and Michal Young. The paper presents a technique to infer relationships between preconditions and postconditions (e.g. postcondition of f implies precondition of g) without knowing the exact invariants. They show that these inferred relationships can improve the quality of invariants produced by Daikon with some manual effort.